Purify let the following message through and even after trying to teach the Jeffrian filter that it was bad it still insists that it is a good e-mail!!


Return-Path: <racquel.Hlady@activemetaphors.net>
Delivered-To: *****@macmail.com
Received: (qmail 18362 invoked by uid 0); 14 Nov 2007 18:44:58 -0000
Received: from unknown (HELO AAmiens-157-1-123-181.w86-215.abo.wanadoo.fr) (86.215.160.129)
by mail.macmail.com with SMTP; 14 Nov 2007 18:44:58 -0000
Received: from evasdream ([189.157.120.147]:32217 "EHLO evasdream"
smtp-auth: <none> TLS-CIPHER: <none> TLS-PEER-CN1: <none>)
by AAmiens-157-1-123-181.w86-215.abo.wanadoo.fr with ESMTP id S22XZOHTRAYETUDC (ORCPT
<rfc822;*****%macmail.com@mail.macmail.com>);
Wed, 14 Nov 2007 19:45:29 +0100

<snip JKH>

Training is not a "one shot" deal.

If you train a message as "bad", its contents are added to the statistics pool for rating the next message. e.g. I have trained a few hundred messages, and have over 10,000 words in my bad words, and ~8000 words in my good words. My filter catches nearly everything.

If you do want a "one shot" capability, you should work your blacklist.
If you put this spammer's domain in your blacklist, you'll never get another email from them.

sentencecame.com

Thanks Jeff,

Looks like I'm about half way there, I've got about 4500 words in each list.

5455 in my good list but only 1715 in my bad. Looks like I've got a lot more training to do.

Can anyone say roughly how long before I can turn off training mode?

Took me about a week. Between country filter, blacklist, and jeffrian filter, spam getting through is almost non-existent.

I'll knock it off in a few days then. I think all I need to do is a bit of tweaking anyway then all should be well.

Just a quick thought, if I turn off training and then a spam email gets through is their any way of training Purify using the offending message afterwards?

I'm thinking about switching it off now as no spam or replies/bounces, etc have got through today.

Wonderful.

Either you are mad or a masochist I know what you mean though, how do you know it's working if it's not producing anything visible The only way I can see would be to e-mail the originator and ask them to spam you again

I can't believe it. Turned training off and then first run one slipped through. I've just gone a couple of days of nothing but genuine emails reaching my inbox.
Unfortunately as I was saying now there is no way of training Purify with this message that I can see.
I've now switched training back on. I suppose all I can do is leave it on and just train messages that need training.

What would happen if you manually added what appeared to be significant "bad" words from this message?

You can manually add items to your Jeffrian filter. Remember, that this does not guarantee that every message containing the new items would be a spam hit. New messages would be rated based on the probability produced by the added items.

If you really want a "one shot and it's spam", I'd use the blacklist.

Once a week or so, I have to make an addition or two to my blacklist. e.g. this week, it's the "New Years" spam that I'm getting requiring an update (reg ex = new\s+year ).

BTW, I've had my Purify installation running on one of my Linux servers here without a hitch for weeks. Based on my filter settings, it's doing a magnificent job filtering my email!